Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Decoding Stability on Causal and Downstream Robustness Ablation Suite Averaged over 4 models
Loading...
0.8
Decoding Δ%
HETA
0.664
1.582
2.5
3.418
Apr 14, 2026
Decoding Δ%
Updated 3d ago
Evaluation Results
Method
Method
Links
Decoding Δ%
HETA
Method Variant=Full
2026.04
0.8
HETA
Method Variant=LR+WIN
2026.04
0.9
HETA
Method Variant=w/o Hes...
2026.04
1.6
HETA
Method Variant=w/o KL
2026.04
1.8
HETA
Method Variant=w/o Tra...
2026.04
2.1
ReAGent
Method Variant=Standard
2026.04
2.4
SEA-CoT
Method Variant=Standard
2026.04
2.5
fAML
Method Variant=Standard
2026.04
2.6
Progressive Inference
Method Variant=Standard
2026.04
2.7
ContextCite
Method Variant=Standard
2026.04
2.9
TDD-backward
Method Variant=Standard
2026.04
3
Peering (PML)
Method Variant=Standard
2026.04
3.1
Integrated Gradients
Method Variant=Standard
2026.04
3.4
Attention Rollout
Method Variant=Standard
2026.04
4.2
Feedback
Search any
task
Search any
task